Instructions to use WaveCut/Cosmos3-Super-Text2Image-Quanto-FP8-Transformer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use WaveCut/Cosmos3-Super-Text2Image-Quanto-FP8-Transformer with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("WaveCut/Cosmos3-Super-Text2Image-Quanto-FP8-Transformer", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Clarify Quanto FP8 quantization
Browse files
README.md
CHANGED
|
@@ -7,15 +7,18 @@ tags:
|
|
| 7 |
- diffusers
|
| 8 |
- fp8
|
| 9 |
- quanto
|
|
|
|
| 10 |
- text-to-image
|
| 11 |
license: other
|
| 12 |
license_name: openmdw1.1-license
|
| 13 |
license_link: https://openmdw.ai/license/1-1/
|
| 14 |
---
|
| 15 |
|
| 16 |
-
# Cosmos3-Super-Text2Image FP8 Transformer
|
| 17 |
|
| 18 |
-
This repository contains a transformer-only FP8/float8
|
|
|
|
|
|
|
| 19 |
|
| 20 |
Read NVIDIA's card, license, safety notes, and prompt-format guidance here:
|
| 21 |
[nvidia/Cosmos3-Super-Text2Image](https://huggingface.co/nvidia/Cosmos3-Super-Text2Image).
|
|
@@ -31,7 +34,7 @@ from diffusers import Cosmos3OmniPipeline, Cosmos3OmniTransformer
|
|
| 31 |
from diffusers.schedulers.scheduling_unipc_multistep import UniPCMultistepScheduler
|
| 32 |
|
| 33 |
transformer = Cosmos3OmniTransformer.from_pretrained(
|
| 34 |
-
"WaveCut/Cosmos3-Super-Text2Image-FP8-Transformer",
|
| 35 |
subfolder="transformer",
|
| 36 |
torch_dtype=torch.bfloat16,
|
| 37 |
)
|
|
|
|
| 7 |
- diffusers
|
| 8 |
- fp8
|
| 9 |
- quanto
|
| 10 |
+
- optimum-quanto
|
| 11 |
- text-to-image
|
| 12 |
license: other
|
| 13 |
license_name: openmdw1.1-license
|
| 14 |
license_link: https://openmdw.ai/license/1-1/
|
| 15 |
---
|
| 16 |
|
| 17 |
+
# Cosmos3-Super-Text2Image Quanto FP8 Transformer
|
| 18 |
|
| 19 |
+
This repository contains a transformer-only FP8/float8 quantization made with Hugging Face Optimum Quanto for [nvidia/Cosmos3-Super-Text2Image](https://huggingface.co/nvidia/Cosmos3-Super-Text2Image).
|
| 20 |
+
|
| 21 |
+
**This is a Quanto quantization, not an NVIDIA ModelOpt/NVFP quantization.** The separate NVFP experiments should be compared against this repo explicitly as a different quantization backend.
|
| 22 |
|
| 23 |
Read NVIDIA's card, license, safety notes, and prompt-format guidance here:
|
| 24 |
[nvidia/Cosmos3-Super-Text2Image](https://huggingface.co/nvidia/Cosmos3-Super-Text2Image).
|
|
|
|
| 34 |
from diffusers.schedulers.scheduling_unipc_multistep import UniPCMultistepScheduler
|
| 35 |
|
| 36 |
transformer = Cosmos3OmniTransformer.from_pretrained(
|
| 37 |
+
"WaveCut/Cosmos3-Super-Text2Image-Quanto-FP8-Transformer",
|
| 38 |
subfolder="transformer",
|
| 39 |
torch_dtype=torch.bfloat16,
|
| 40 |
)
|